[Bugfix][MoE] Restore routed output unpadding before shared expert add (#45707)

Signed-off-by: Netanel Haber <58652339+netanel-haber@users.noreply.github.com>
Co-authored-by: mergify[bot] <37929162+mergify[bot]@users.noreply.github.com>
This commit is contained in:
Netanel Haber
2026-06-16 16:06:28 +03:00
committed by GitHub
co-authored by mergify[bot] <37929162+mergify[bot]@users.noreply.github.com>
parent a8c86eeb16
commit c5e5c33fcd
2 changed files with 13 additions and 18 deletions
+2 -9
View File
@@ -283,7 +283,7 @@ def test_nemotron_fp8_moe_vllm_triton(monkeypatch: pytest.MonkeyPatch):
)
def test_nemotron_fp4_moe_flashinfer_throughput(monkeypatch: pytest.MonkeyPatch):
def test_nemotron_fp4_moe_flashinfer_cutlass(monkeypatch: pytest.MonkeyPatch):
can_initialize(
"nvidia/NVIDIA-Nemotron-3-Nano-30B-A3B-NVFP4",
hf_overrides=HF_OVERRIDE_TEXT,
@@ -291,14 +291,7 @@ def test_nemotron_fp4_moe_flashinfer_throughput(monkeypatch: pytest.MonkeyPatch)
)
@pytest.mark.skip(
reason=(
"FP4 MoE backend FLASHINFER_TRTLLM does not support the "
"deployment configuration since kernel does not support "
"hidden_dim % 512 != 0."
)
)
def test_nemotron_fp4_moe_flashinfer_latency(monkeypatch: pytest.MonkeyPatch):
def test_nemotron_fp4_moe_flashinfer_trtllm(monkeypatch: pytest.MonkeyPatch):
can_initialize(
"nvidia/NVIDIA-Nemotron-3-Nano-30B-A3B-NVFP4",
hf_overrides=HF_OVERRIDE_TEXT,